Student Engagement Recognition in Online Courses Based on Transfer Learning

The rapid expansion of online courses has brought significant advantages, but it also exposes several challenges, particularly in terms of inadequate supervision and delayed feedback. These issues prevent teachers from recognizing the difficulties faced by underperforming students, exacerbating the disparity between those who understand the material and those who struggle. As a result of delayed assistance from teachers, the students who struggle are more likely to experience declining grades and, in severe cases, drop out of courses. To address these challenges, the author introduces a deep learning-based engagement recognition model that utilizes computer vision technology to monitor student engagement in online courses. The model is evaluated using the DAiSEE open-source dataset, which consists of videos from real-world online learning environments. Given the current lack of student engagement datasets and the ambiguity in engagement categorization within existing datasets, researchers often face challenges in finding suitable datasets for training. To overcome this obstacle, the author applies transfer learning by utilizing benchmark emotion datasets to represent students' learning emotions. This innovative approach provides a valuable training strategy for future research. It offers an alternative for researchers who may lack access to large-scale, accurately labeled student images, making it a promising solution for the development of student engagement recognition models in the future.